AI chatbots from Anthropic, OpenAI and Google can now search the web on a user's behalf, and WIRED reports that the feature is genuinely useful in some situations and risky in others. In a guide published on 2 November 2026, the outlet set out which kinds of queries suit an AI search and which are better handled the old-fashioned way. The distinction, according to WIRED, comes down to how well-known the subject is and how much precision the answer demands.
Why it matters
Searching through a chatbot is convenient, but it shifts several decisions away from the user. The AI has to interpret the question, choose which pages to consult and decide what to extract from them. WIRED notes that this saves time while making the process less transparent, which matters most when an answer could affect someone's health, money or legal position. Understanding where the tool is reliable helps users avoid treating a confident-sounding reply as a verified fact.
How AI search actually works
When asked to look something up, the newest versions of these chatbots blend two things, according to WIRED. They run what amounts to an ordinary web search, then combine the results with material drawn from their training data. The output arrives in the polished, conversational format typical of generative AI.
That added layer is where the behaviour differs from a plain list of links. WIRED reports that the models are strongest at paraphrasing, summarising and combining material, and that these same processes can introduce errors along the way.
Where it performs well
For simple, widely documented facts, WIRED considers AI search a safe choice. The example given is the height of the Empire State Building: information like this is easy to find online, and the probability patterns that large language models rely on make it highly likely the right figure surfaces.
The tool also shines when a question is too complicated for a standard search box. WIRED cites a request for family-friendly San Francisco restaurants with a $200 budget. An AI can pull from several sources at once and produce a single synthesised answer, whereas a conventional search would return links ranked on simpler criteria. The outlet points out that the bots have never eaten in San Francisco; they are drawing on public write-ups by people who have.
WIRED also references Stanford University research indicating that when an AI locates an accurate source, it is very good at extracting the correct detail from it. The failures tend to occur when no accurate source exists, or when the query is misread, and the system still delivers a confident answer.
Where traditional searching still wins
Obscure questions are the weak spot. WIRED gives the example of identifying the fifth-billed actor in a little-known television programme. With only one or two references online, the AI may pick the wrong one.
Precision is another dividing line. For an exact quotation, a figure from a report or a device specification, WIRED suggests going to Google directly, because summarising and paraphrasing can distort fine detail.
Control over sources is a third. Someone who knows which outlets they trust for technology coverage can favour those sites in a conventional search, something an AI may not replicate. WIRED cites researchers at UC San Diego who found that AI web searches sometimes select sources for how closely they match the query rather than for their authority. A personal blog could therefore be chosen over a professional medical site. A person scanning a list of results would more likely pick the source they trust.
Breaking news is a further gap, since the data an AI retrieves can lag behind events. WIRED also advises against relying on AI search for high-stakes questions involving health, the law or personal finances, where a mistake carries real consequences.
Checking the answer
WIRED's guidance is to verify any AI search result to some degree. When an AI-assisted approach is warranted, users should double- and triple-check the sources the chatbot cites, which should appear as links within the answer. If anything is unclear, the user can ask the bot where a claim came from, which helps establish whether the model produced it rather than found it.
Traditional search results can also contain false information, WIRED acknowledges, but the reader is in a better position to judge what is accurate when the sources are laid out for them.
The practical takeaway from the guide is that AI search suits common questions with plenty of current, reliable coverage online, and that conventional searching remains the safer route for niche topics, exact details, breaking stories and decisions that carry risk. Watch for further research on how these systems rank sources, and for changes to how chatbots display the links behind their answers.
